discrete transistors

From: Allison J Parent <allisonp_at_world.std.com>
Date: Tue Oct 20 11:00:53 1998

< I expect this was almost essential; all of the wiring between circuit
< cards was on the inside of the circle, and the customer engineers had to
< able to get to that wiring.

True.

< I suspect that the reason Seymour Cray built the machine in the shape h
< did was that the circuit cards, plus the machined columns that supporte
< and cooled them, were wider than the card connectors. By arranging the
< card columns in a semi-circle with the connectors on the inside, he coul
< minimize wiring length.

It was a design requirement that no signal path would exceed 1 meter.
It also allowed signals like clock to be distributed in a way that
assured all parts were geting the same drumbeat at the same time. If it
were a linear layout it would easily be several meters long and the
signals from either end would have cables long enough to insure the
data was way late.

Signal propagation in cables is less than speed of light (c) and can be
as low as 75% in some coax style cables. So if you have a 3 meter cable
and you putting pulses out at the rate of one every 3nS you can literally
have three pulses traveling down the wire like golf balls in a tube! If
that pulse is your sychronizing clock you can see that when it gets to
the end of that 3meter wire it's late (by two pulses) compared to the
logic at the beginning of the wire. This is a handicap for ultrafast
systems but it can be useful for storing data too (delay line memories
and timing elements)! So this is why big and fast do not go well
together.

It also was an advantageous layout for cooling and power distribution.
It made for an unusual looking machine with a lot going on down under
the subfloor! Customer engineers were not that fond of it as it tended
to be cold in the middle!

This phenomenon is seen and known at the chip level and influences
how the logic is organized on the die for super fast logic.

Allison
Received on Tue Oct 20 1998 - 11:00:53 BST

This archive was generated by hypermail 2.3.0 : Fri Oct 10 2014 - 23:31:27 BST